OpenAI has introduced an Admin plugin for ChatGPT Work and Codex, giving workspace administrators a way to review usage, manage permissions, and approve spending requests directly inside a chat conversation instead of navigating a separate console. The company announced the plugin on August 25, 2026, describing it as a faster way to move from a question to an action without switching between tools.
The launch matters less because it introduces new capabilities and more because of what it consolidates. As organizations lean harder on ChatGPT Work and Codex for daily operations, the administrative overhead of tracking usage, managing access, and approving requests has grown alongside it. OpenAI’s answer is not a new permissions system, but a conversational layer over the tools admins already have. The launch was independently reported the same day by outlets including 9to5Mac.
What OpenAI Announced
According to OpenAI’s official announcement, the Admin plugin is installable from the Plugins directory inside ChatGPT Work and Codex. Once enabled by a workspace admin and installed from the directory, it allows admins to:
- Review activity and credit usage across ChatGPT Work and Codex, including when members or groups are approaching credit limits
- Add or remove members, update groups, and handle onboarding and offboarding
- Review effective permissions, diagnose access issues, and control feature or model access by role or group
- Adjust usage limits and approve or deny spending requests with context
OpenAI also says the plugin can automate recurring administrative work without custom engineering — for example, routing pending usage requests to Slack or Microsoft Teams, where an authorized reviewer can approve or deny them within tools they already use.
It’s worth being precise about what this is not. It is not a new admin console, and it does not expand what any individual admin is allowed to do. It’s a new way to reach capabilities that already existed.
How the Admin Plugin Works
OpenAI describes the plugin’s mechanism in fairly specific terms: it maps an admin’s natural-language instructions to a supported read or write action, then returns a structured result showing what was requested, whether it completed, and what changed. For actions with broader impact, admins can review the change before it’s applied.
Crucially, the plugin “works within each user’s existing role and permissions” and “does not grant broader access,” according to OpenAI. In practice, that means the plugin behaves like a conversational front end — it doesn’t sit above or outside the workspace’s existing access-control rules, it operates through them.
What OpenAI has not disclosed is the underlying technical implementation: no specific API endpoint, protocol, or architecture has been published. That detail simply isn’t public yet.
ChatGPT Work and Codex — What They Actually Are
Some coverage of AI product launches blurs naming details, so it’s worth being clear here. ChatGPT Work is not a new subscription tier, and it isn’t a renamed version of an older plan. It’s a mode within ChatGPT — sitting alongside standard Chat and the Codex coding surface — that launched on July 9, 2026 alongside the GPT-5.6 model family. Rather than working through a conversation message by message, Work takes a described outcome, pulls in context from connected tools, and works through multi-step tasks over an extended period before returning a finished document, spreadsheet, or similar deliverable. It’s bundled into existing Plus, Pro, Business, and Enterprise plans, with usage drawn from each plan’s existing allowance rather than priced separately.
Codex, in this context, refers to OpenAI’s current agentic coding surface inside ChatGPT — not the original Codex API model that OpenAI deprecated in 2023. The name has persisted, but the product underneath it today is a different, actively developed tool built for software development work, distinct from Work’s more general-purpose focus.
| Term | What It Is |
|---|---|
| ChatGPT Work | A mode within ChatGPT for multi-step, outcome-based tasks, launched July 9, 2026 |
| Codex | OpenAI’s current agentic coding surface — distinct from the earlier Codex API model deprecated in 2023 |
| Global Admin Console | A separate console for cross-workspace analytics, Codex usage data, and billing |
| Admin plugin | A conversational plugin inside ChatGPT Work and Codex for usage, permissions, and spend management, announced August 25, 2026 |
Relationship to OpenAI’s Existing Admin Tools
The Admin plugin sits alongside, not in place of, OpenAI’s Global Admin Console, which already provides cross-workspace analytics, a dedicated Codex usage view, and billing details for Enterprise and Edu workspaces. Core settings — Members, Groups, Roles and Permissions, GPTs, Skills, and Plugins — are managed separately in ChatGPT’s Workspace settings. The plugin appears to expose that same underlying data and those same actions conversationally, rather than adding functionality the console lacks.
Availability and Rollout
Getting started requires a workspace admin to first enable the plugin in workspace settings, then install it from the Plugins directory in ChatGPT Work on web or desktop.
One detail matters for larger organizations: in Enterprise and Edu workspaces, plugins and their underlying apps are disabled by default. Admins must explicitly choose to make a plugin available or installed for eligible roles. That means the Admin plugin won’t simply appear for every enterprise customer — it requires deliberate opt-in, which is consistent with how OpenAI treats other plugins in security-sensitive workspace tiers.
OpenAI has not published specific regional restrictions for the Admin plugin itself, so it isn’t confirmed whether availability differs by geography.
Security and Permissions Model
The plugin inherits its access boundaries from OpenAI’s existing role-based access control (RBAC) system, which governs permissions across both the ChatGPT dashboard and the API, layered under workspace identity controls like single sign-on and, in some tiers, SCIM-based group sync.
This matters because plugins and connected apps are widely treated as a security consideration industry-wide, since external integrations can create new data pathways — which is likely why OpenAI keeps plugins off by default in Enterprise and Edu workspaces and lets admins restrict them by role. OpenAI has not disclosed the specific authentication protocol behind the Admin plugin itself.
OpenAI’s Internal Use Case
OpenAI’s announcement includes a description of how its own IT team is using the plugin and the broader ChatGPT Work and Codex environment. Kunal Malik, OpenAI’s Head of Global IT, said the plugin “connects the question to the next supported action” — whether that’s checking permissions, updating a group, changing a usage limit, or reviewing a spending request — while preserving the team’s existing controls.
OpenAI also states that a ChatGPT Work agent deployed in Slack handles employee IT requests and, at the time of reporting, resolved roughly 45 percent of ticket volume. The company says turning support-ticket data into operational dashboards helped its IT team eliminate a backlog even as support volume roughly doubled.
These figures are worth reading with the right context: they’re OpenAI’s own internal, company-reported numbers, not independently audited or verified by a third party. They describe OpenAI’s use of its own product, not a customer deployment.
How It Compares to Other Enterprise AI Admin Tools
OpenAI isn’t alone in building governance tools around enterprise AI deployment. Microsoft manages access to its AI services through Azure role-based access control, which lets organizations assign different permission levels to team members across Azure resources. Anthropic’s Claude for Teams offers single sign-on, role-based access, and admin-level audit logs as part of its enterprise controls.
What isn’t confirmed is whether any competitor currently offers a directly comparable conversational admin plugin — a tool that lets administrators query usage data and take supported actions through natural-language chat rather than a traditional console interface. No equivalent product from Microsoft, Google, or Anthropic surfaced in available research. That absence of evidence doesn’t mean OpenAI is first to market with this approach; it means a like-for-like comparison isn’t currently available.
Limitations and Open Questions
Several details remain unresolved. There’s no independent, third-party data yet on how well the plugin performs in practice — all current figures come from OpenAI itself, and the plugin launched too recently for outside evaluation to exist. It also isn’t confirmed how the Admin plugin interacts with Enterprise Key Management, the encryption feature that lets organizations control their own decryption keys, since related workspace-agent features have been noted as unavailable in EKM-enabled workspaces at launch.
The exact technical implementation — what API or protocol underlies the plugin’s read and write actions — has not been published. And while general plugin availability elsewhere depends on plan, role, and region, no region-specific restriction has been confirmed for the Admin plugin specifically.
Timeline at a Glance
- December 2025: OpenAI announces Enterprise Key Management (EKM) for ChatGPT Enterprise and Edu workspaces.
- July 9, 2026: GPT-5.6 and ChatGPT Work launch; OpenAI migrates its App directory into the Plugin directory.
- August 25, 2026: OpenAI announces the Admin plugin for ChatGPT Work and Codex.
Conclusion
The Admin plugin doesn’t change what a ChatGPT Work or Codex administrator is allowed to do — it changes how they do it, replacing some console navigation with a conversation. For organizations already managing sizable ChatGPT deployments, that could meaningfully cut down on the friction of routine tasks like approving spend requests or auditing permissions.
What’s genuinely uncertain right now is how well this holds up outside OpenAI’s own IT team, since the only performance data available so far comes from the company itself. Whether the plugin extends smoothly into EKM-protected workspaces, whether OpenAI adds region-specific rollout details, and whether competitors respond with something comparable are all open questions worth watching as adoption widens beyond this initial announcement.
Frequently Asked Questions
What is the OpenAI Admin plugin for ChatGPT Work and Codex? It’s a plugin, installable from the Plugins directory, that lets workspace administrators review usage and credit data, manage members and groups, review permissions, and approve or deny spending requests directly within a ChatGPT Work or Codex conversation.
Is ChatGPT Work a new ChatGPT subscription tier? No. ChatGPT Work is a mode within ChatGPT — alongside standard Chat and Codex — launched July 9, 2026. It’s bundled into existing Plus, Pro, Business, and Enterprise plans rather than sold as its own tier.
Does the Admin plugin give admins new permissions? No. OpenAI states explicitly that the plugin operates within each user’s existing role and permissions and does not grant broader access than an admin already has.
How do I enable the Admin plugin in my workspace? A workspace admin first enables the plugin in ChatGPT workspace settings, then installs it from the Plugins directory in ChatGPT Work on web or desktop.
Is the Admin plugin available in Enterprise and Edu workspaces by default? No. Plugins and their underlying apps are disabled by default in Enterprise and Edu workspaces, so an admin must explicitly enable and install the Admin plugin for eligible roles.
How does the Admin plugin relate to OpenAI’s Global Admin Console? The plugin appears to expose data and actions that already exist across the Global Admin Console and ChatGPT Workspace settings, delivered conversationally rather than through console navigation. It does not replace the console, and core settings like Members, Groups, and Roles remain managed in ChatGPT Workspace settings.
Is “Codex” in this plugin the same as the older Codex model OpenAI deprecated? No. The Codex referenced here is OpenAI’s current, actively developed agentic coding surface inside ChatGPT. It’s unrelated to the earlier Codex API model that OpenAI deprecated in 2023, beyond sharing the same name.